Merge pull request #7 from srivarra/exposure/adjust_sigmoid

✨ Added  Adjust Sigmoid

docs/api.md

@@ -11,4 +11,5 @@
 
     exposure.adjust_gamma
     exposure.adjust_log
+    exposure.adjust_sigmoid
 ```

docs/notebooks/example.ipynb

@@ -117,6 +117,24 @@
     "an.exposure.adjust_gamma(blobs_image, gamma=1.5, gain=2.0).pipe(an.exposure.adjust_log, gain=2.0).compute()"
    ]
   },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "se.adjust_sigmoid(blobs_image, cutoff=0.5, gain=10, inv=True)"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "an.exposure.adjust_sigmoid(blobs_image, cutoff=0.5, gain=10, inv=True).compute()"
+   ]
+  },
   {
    "cell_type": "code",
    "execution_count": null,

src/anatomize/exposure/__init__.py

@@ -1,3 +1,3 @@
-from .basic import adjust_gamma, adjust_log
+from .basic import adjust_gamma, adjust_log, adjust_sigmoid
 
-__all__ = ["adjust_gamma", "adjust_log"]
+__all__ = ["adjust_gamma", "adjust_log", "adjust_sigmoid"]

src/anatomize/exposure/_utils.py

@@ -7,7 +7,9 @@
 
 
 def _rechunk(image: xr.DataArray, chunks: Mapping[str, int] | None) -> xr.DataArray:
-    _, n_y, n_x = image.shape
+    n_c, n_y, n_x = image.shape
+    if image.chunks == (1, n_y, n_x):
+        return image
     if chunks is None:
         chunks = {C: 1, Y: n_y, X: n_x}
     return image.chunk(chunks)
@@ -68,3 +70,25 @@ def _log(Ic, gain, out):
 )
 def _inv_log(Ic, gain, out):
     out[:] = gain[:] * (2 ** Ic[:] - 1)
+
+
+@guvectorize(
+    ["void(float64[:,:], float64[:], float64[:], float64[:,:])"],
+    "(m,n),(c),(c)->(m,n)",
+    nopython=True,
+    cache=True,
+    target="cpu",
+)
+def _sigmoid(Ic, cutoff, gain, out):
+    out[:] = 1 / (1 + np.exp(gain[:] * (cutoff[:] - Ic[:])))
+
+
+@guvectorize(
+    ["void(float64[:,:], float64[:], float64[:], float64[:,:])"],
+    "(m,n),(c),(c)->(m,n)",
+    nopython=True,
+    cache=True,
+    target="cpu",
+)
+def _inv_sigmoid(Ic, cutoff, gain, out):
+    out[:] = 1 - (1 / (1 + np.exp(gain[:] * (cutoff[:] - Ic[:]))))

src/anatomize/exposure/basic.py

@@ -3,7 +3,7 @@
 from spatialdata.models import C, X, Y
 
 from anatomize.core.decorators import convert_kwargs_to_xr_vec
-from anatomize.exposure._utils import _gamma, _inv_log, _log, _normalize, _rechunk
+from anatomize.exposure._utils import _gamma, _inv_log, _inv_sigmoid, _log, _normalize, _rechunk, _sigmoid
 
 
 def iter_channels(image: xr.DataArray) -> xb.BatchGenerator:  # noqa: D103
@@ -122,3 +122,40 @@ def adjust_log(image: xr.DataArray, gain: float = 1, inv=False) -> xr.DataArray:
         output_dtypes=[data.dtype],
         dask_gufunc_kwargs={"allow_rechunk": True},
     )
+
+
+@convert_kwargs_to_xr_vec("cutoff", "gain")
+def adjust_sigmoid(image: xr.DataArray, cutoff: float = 0.5, gain: float = 10, inv=False) -> xr.DataArray:
+    """Performs Sigmoid Correction on the input image.
+
+    Parameters
+    ----------
+    image
+        The image to adjust.
+    cutoff
+        Cutoff of the sigmoid function that shifts the characteristic curve in horizontal direction, by default 0.5
+    gain
+        The constant multiplier in exponential's power of sigmoid function, by default 10
+    inv
+        If True, the negative sigmoid correction is performed, otherwise
+        the sigmoid correction is performed, by default False.
+
+    Returns
+    -------
+    The sigmoid adjusted image.
+    """
+    data = _rechunk(image, chunks=None)
+
+    f = _sigmoid if not inv else _inv_sigmoid
+
+    return xr.apply_ufunc(
+        f,
+        data,
+        cutoff,
+        gain,
+        input_core_dims=[[Y, X], [], []],
+        output_core_dims=[[Y, X]],
+        dask="parallelized",
+        output_dtypes=[data.dtype],
+        dask_gufunc_kwargs={"allow_rechunk": True},
+    )

tests/exposure/test_basic.py

@@ -34,3 +34,19 @@ def test_adjust_log(blobs_sdata_store: UPath, gain: float, inv: bool):
         dims=blobs_sdata.images["blobs_image"].dims,
     )
     xr.testing.assert_allclose(a=blobs_image_log_xr, b=blobs_image_log_skimage, atol=1e-10)
+
+
+@pytest.mark.parametrize("cutoff,gain,inv", ([0.5, 10, True], [2.0, 20, False]))
+def test_adjust_sigmoid(blobs_sdata_store: UPath, cutoff: float, gain: float, inv: bool):
+    blobs_sdata = sd.read_zarr(blobs_sdata_store)
+
+    blobs_image_sigmoid_xr = an.exposure.adjust_sigmoid(
+        image=blobs_sdata.images["blobs_image"], cutoff=cutoff, gain=gain, inv=inv
+    )
+
+    blobs_image_sigmoid_skimage = xr.DataArray(
+        data=exposure.adjust_sigmoid(blobs_sdata.images["blobs_image"], cutoff=cutoff, gain=gain, inv=inv),
+        coords=blobs_sdata.images["blobs_image"].coords,
+        dims=blobs_sdata.images["blobs_image"].dims,
+    )
+    xr.testing.assert_allclose(a=blobs_image_sigmoid_xr, b=blobs_image_sigmoid_skimage, atol=1e-10)